Notes for May 6-12

Thanks to the lethargy brought upon by allergies and the beginning of the warm season, this was a week where most of my free time was spent fixing things and building tools.

The Third Llama

Since the ollama instance on my has turned out to be fast enough to run llama3 for most practical uses (even though the RTX 3060 on is much faster), I decided to start stacking things on top. But there’s a functional gap between Open Web UI and my existing Python tooling that I’d like to fix, because I spend too much time futzing around with API calling.

After playing around with Dify the other day, I decided to see how hard it would be to duplicate its core functionality with , just because I’d rather invest in a stack I fully control. After much less than an hour, I had enough re-usable building blocks to do this:

A trivial, almost meaningless use of intellectual capacity on all counts.

RAG is also easy enough to do, although I’m still waiting for the ArchiveBox API to come out in the next release–so right now I’m just playing with plain web searches while I build out the rest of the tools:

I need to have some fun while I'm doing this

ReAct (which is so badly named that it’s almost impossible to search for) also seems like a possibility, although the key challenge there is going to be looping and prompt chaining.

Other Concerns

Other than that, my free time was eaten up by a few outings (including real-life trips to the local office, which provided a long overdue reminder of the mixed blessings of commuting), a bit of exercise (also long overdue, and still too little) and reading.

Even as my hardware project backlog continues to mount (I have several things to solve, from NAS storage limitations to HID stuff, plus of course pending hardware testing), I’m trying to slowly get into the Summer break mood and the inevitable slowdown that comes with it.

I’ve been itching to do more electronics, though. The most satisfactory thing I did all week was replacing an old iPhone 5/SE’s broken screen with a cheap Chinese part (including transplanting the Touch ID button to the new front panel), which only highlighted how much better that form factor was to hold in the hand when compared to today’s monstrosities.